вход по аккаунту



код для вставкиСкачать
Nuclear Inst. and Methods in Physics Research, A 904 (2018) 44–63
Contents lists available at ScienceDirect
Nuclear Inst. and Methods in Physics Research, A
journal homepage:
Characterisation and testing of CHEC-M—A camera prototype for the
small-sized telescopes of the Cherenkov telescope array
J. Zorn a, *, R. White a, *, J.J. Watson b, *, T.P. Armstrong b,c , A. Balzer d , M. Barcelo a , D. Berge d,1 ,
R. Bose e , A.M. Brown c , M. Bryan d , P.M. Chadwick c , P. Clark c , H. Costantini f , G. Cotter b ,
L. Dangeon g , M. Daniel h,2 , A. De Franco b , P. Deiml i , G. Fasola g , S. Funk i,j , M. Gebyehu d ,
J. Gironnet k , J.A. Graham c , T. Greenshaw h , J.A. Hinton a , M. Kraus i , J.S. Lapington l ,
P. Laporte g , S.A. Leach l , O. Le Blanc g , A. Malouf m , P. Molyneux l , P. Moore e,3 , H. Prokoph d,1 ,
A. Okumura n , D. Ross l , G. Rowell m , L. Sapozhnikov j , H. Schoorlemmer a , H. Sol g ,
M. Stephan d , H. Tajima n , L. Tibaldo a,4 , G. Varner o , A. Zink i
Max-Planck-Institut für Kernphysik, P.O. Box 103980, 69029 Heidelberg, Germany
Department of Physics, University of Oxford, Keble Road, Oxford OX1 3RH, UK
Department of Physics and Centre for Advanced Instrumentation, Durham University, South Road, Durham DH1 3LE, UK
GRAPPA, University of Amsterdam, Science Park 904, 1098 XH Amsterdam, The Netherlands
Department of Physics, Washington University, St. Louis, MO 63130, USA
f Aix Marseille Université, CNRS/IN2P3, CPPM, 163 avenue de Luminy, case 902, 13288 Marseille, France
g Observatoire de Paris, CNRS, PSL University, LUTH & GEPI, Place J. Janssen, 92195, Meudon cedex, France
h University of Liverpool, Oliver Lodge Laboratory, P.O. Box 147, Oxford Street, Liverpool L69 3BX, UK
Erlangen Centre for Astroparticle Physics (ECAP), Erwin-Rommel-Str. 1, D 91058 Erlangen, Germany
Kavli Institute for Particle Astrophysics and Cosmology, Department of Physics and SLAC National Accelerator Laboratory, Stanford University, 2575 Sand Hill Road,
Menlo Park, CA 94025, USA
CNRS, Division technique DT-INSU, 1 Place Aristide Briand, 92190 Meudon, France
Department of Physics and Astronomy, University of Leicester, University Road, Leicester, LE1 7RH, UK
m School of Physical Sciences, University of Adelaide, Adelaide5005, Australia
n Institute for Space–Earth Environmental Research, Nagoya University, Furo-cho, Chikusa-ku, Nagoya, Aichi 464-8601, Japan
o University of Hawai’i at Manoa, 2500 Campus Rd, Honolulu, HI 96822, USA
Imaging atmospheric Cherenkov telescopes
Cherenkov telescope array
Full-waveform readout
The Compact High Energy Camera (CHEC) is a camera design for the Small-Sized Telescopes (SSTs; 4 m diameter
mirror) of the Cherenkov Telescope Array (CTA). The SSTs are focused on very-high-energy -ray detection via
atmospheric Cherenkov light detection over a very large area. This implies many individual units and hence
cost-effective implementation, as well as shower detection at large impact distance, and hence large field of view
(FoV), and efficient image capture in the presence of large time gradients in the shower image detected by the
camera. CHEC relies on dual-mirror optics to reduce the plate-scale and make use of 6 × 6 mm2 pixels, leading
to a low-cost (∼150 ke), compact (0.5 m × 0.5 m), and light (∼45 kg) camera with 2048 pixels providing a
camera FoV of ∼9 degrees. The CHEC electronics are based on custom TARGET (TeV array readout with GSa/s
sampling and event trigger) application-specific integrated circuits (ASICs) and field programmable gate arrays
(FPGAs) sampling incoming signals at a gigasample per second, with flexible camera-level triggering within a
single backplane FPGA. CHEC is designed to observe in the -ray energy range of 1–300 TeV, and at impact
distances up to ∼500 m. To accommodate this and provide full flexibility for later data analysis, full waveforms
with 96 samples for all 2048 pixels can be read out at rates up to ∼900 Hz. The first prototype, CHEC-M, based on
Corresponding authors.
E-mail addresses: (J. Zorn), (R. White), (J.J. Watson).
Now at: Deutsches Elektronen-Synchrotron, Platanenallee 6, 15738 Zeuthen, Germany.
Now at: Harvard-Smithsonian Center for Astrophysics, 60 Garden Street, Cambridge, MA 02138, USA.
Deceased on 20/05/2017.
Now at: Institut de Recherche en Astrophysique et Planétologie, CNRS-INSU, Université Paul Sabatier, 9 avenue Colonel Roche, BP 44346 31028 Toulouse
Cedex 4, France.
Received 12 June 2018; Received in revised form 26 June 2018; Accepted 27 June 2018
Available online 4 July 2018
0168-9002/© 2018 Elsevier B.V. All rights reserved.
J. Zorn et al.
Nuclear Inst. and Methods in Physics Research, A 904 (2018) 44–63
multi-anode photomultipliers (MAPMs) as photosensors, was commissioned and characterised in the laboratory
and during two measurement campaigns on a telescope structure at the Paris Observatory in Meudon. In this
paper, the results and conclusions from the laboratory and on-site testing of CHEC-M are presented. They have
provided essential input on the system design and on operational and data analysis procedures for a camera of
this type. A second full-camera prototype based on Silicon photomultipliers (SiPMs), addressing the drawbacks of
CHEC-M identified during the first prototype phase, has already been built and is currently being commissioned
and tested in the laboratory.
1. Introduction
the same lower energy threshold as in option 1) providing a wide FoV
(e.g. >10◦ , see [8]).
CTA is a multi-km2 array following the former approach with
telescope spacings of 100–200 m. To cover a wide energy range (from
20 GeV to 300 TeV) it uses three telescope classes (referred to as small-,
medium-, and large-sized) with mirror diameters of 4 to 23 m and
different quantities of telescopes of each class. Furthermore, CTA will
consist of a northern and a southern hemisphere site to cover the whole
sky. -rays at TeV energies are few in number but initiate EASs which
produce a high number of Cherenkov photons (compared to -rays at
energies of a few tens of GeV being high in number but initiating a
relatively low number of Cherenkov photons, see e.g. [9,10]). Since they
are expected to mostly origin from Galactic sources, which can be better
observed (at higher elevation) from the southern hemisphere, the southern hemisphere site is envisaged to host 70 small-sized telescopes (SSTs).
For a conventional IACT design with the size of an SST (mirror diameter
of ∼4 m), following single-mirror-dish-design (parabolic reflector or
Davies–Cotton [11] design) such as H.E.S.S., VERITAS, and MAGIC [12–
14], the cost of the camera dominates that of the telescope. The use of a
secondary reflector and aspherical Schwarzschild–Couder (SC) [15,16]
optics enables a reduction in the plate scale of a 9◦ -FoV telescope by
a factor of ∼3 [17]. The reduced plate scale introduces rather novel
options for the camera photosensor technology, including multi-anode
photomultipliers (MAPMs) and Silicon photomultipliers (SiPMs). These
offer considerably reduced cost with respect to photomultiplier tubes
(PMTs) used in cameras of conventional IACTs. Thus, the overall camera
cost is reduced allowing a larger array of SSTs, and therefore increased
area coverage at fixed cost and telescope spacing. Two such dual-mirror
optical designs are currently being prototyped for the SSTs of CTA:
GCT [18,19] and ASTRI [20].
The Compact High Energy Camera (CHEC) is a proposed camera
suitable for use in both of these CTA dual-mirror SST designs and
under development by groups from Australia, Germany, Japan, the
Netherlands, the UK, and the US. The required compact nature of
CHEC hinges on the use of commercially available multi-pixel photoncounting photosensors and the custom TARGET (TeV array readout
with GSa/s sampling and event trigger) application-specific integrated
circuits (ASICs) [21] to provide a high-performance low-cost solution.
Research and development for CHEC is progressing via the development
of two prototype cameras: CHEC-M, based on MAPMs, and CHEC-S,
based on SiPMs. In this paper, we present results from CHEC-M, the first
CHEC prototype. The prototype design and laboratory test results are
presented. On-sky Cherenkov data is shown from CHEC-M installed on a
GCT prototype telescope in Meudon near Paris, and upcoming prospects
towards a camera design for the production phase of CTA are discussed.
The current energy frontier for high-energy astronomy lies at
around 50 TeV. Above this energy existing instruments detect only
a handful of photons. So far, no photons with energies of >100 TeV
have been observed from any source and only a few sources with ray energies above 30 TeV have been detected, e.g. the supernova
remnant RX J1713.7-3946, the pulsar wind nebulae Crab and VelaX, and the extended sources MGROJ2031+41, MGROJ2019+37 and
MGROJ1908+06. Increased collection area is therefore a key requirement for future high-energy -ray instruments. Beside ground-based
extensive air shower (EAS) arrays like HAWC [1] and LHAASO [2],
arrays of imaging atmospheric Cherenkov telescopes (IACTs) with an
area of ∼10 km2 are the most promising candidates for instruments
being able to push the upper energy frontier to higher energies while
still maintaining the necessary level of background suppression and
providing excellent angular resolution [3,4]. Despite the significant
technical challenges and potential cost implications of such a large
area instrument, it remains a highly attractive scientific prospect, for
example in the search for the sources of Galactic cosmic rays up to the
so-called knee in the cosmic-ray spectrum at PeV energies, and the search
for new physics including Lorentz invariance violation and axion-like
particles. When observing with a wide-energy coverage instrument, such
as the Cherenkov Telescope Array (CTA) [3,5], the wide spectral range
makes it possible to remove ambiguities on the nature of the radiating
particles, with inverse Compton emission strongly suppressed by the
Klein–Nishina effect at >100 TeV.
IACTs record the Cherenkov light emitted by the secondary particles
of the EAS initiated by very-high-energy (>100 GeV) -rays or by
charged cosmic rays. Since at these energies cosmic-ray events are
several hundred times (the exact number being dependent on the energy
of the primary particle) more abundant than -ray events, they are the
main background and need to be excluded in the analysis. Showers
imaged by IACTs reach maximum development at heights of ∼10 km
above the observer. Due to the height dependence of the Cherenkov
emission angle in the atmosphere ranging from about 0.8◦ at a height
of 10 km a.s.l. up to a maximum of 1.4◦ at sea level, the radius of the
Cherenkov light pool on the ground at 1500 m a.s.l. is about 120 m
(for vertical showers).5 Assuming the shower axis being parallel to
the telescope axis, the maximum of the observed Cherenkov emission
is displaced by angles of ≈0.8◦ × ( 120 m ) from the telescope axis (using
small-angle approximations, tan  ≈  for angles  ≲ 1◦ ), where  is
the distance of the shower core position on ground to the telescope.
Two approaches therefore exist for the design of a multi-km2 array:
either (1) a large number of closely spaced telescopes with a modest
field of view (FoV) (e.g. as in existing IACTs like H.E.S.S. [6,7] with
an inter-telescope spacing of ∼100 m and a FoV of ∼5◦ ) or (2) wider
inter-telescope spacings (e.g. >800 m) with larger mirrors (to achieve
2. Concept
Above a few TeV, the Cherenkov light intensity is such that showers
can also be detected outside the light pool of fairly uniform illumination
of about 200–250 m diameter. Thus, an energy threshold of around
1 TeV can be achieved with a telescope spacing of ∼250 m and a
telescope diameter  of only ∼4 m. An angular pixel size of ∼0.2◦
is required to be less than the full width half maximum (FWHM) of
a typical 1 TeV -ray image [3,22]. It may then be matched to lowcost photosensors of ∼6 mm diameter, setting the telescope focal length
This can also be roughly calculated using geometry and the emission angle
at 10 km a.s.l.: (10 000 m−1500 m) tan(0.8◦ ) ≈ 120 m.
J. Zorn et al.
Nuclear Inst. and Methods in Physics Research, A 904 (2018) 44–63
Fig. 1. Schematic showing the logical elements of CHEC-M, the communication between those elements, the raw data flow through the camera, the trigger
architecture, and the clock distribution scheme.
Source: Reproduced from [23].
etc.) and taking action if needed. Additionally, hardware allowing insitu gain calibration, flat-fielding measurements, and regular monitoring
of the camera and telescope response is required.
 to ∼2 m. The resulting  ∕ ratio of ∼0.5 can be achieved using a
dual-mirror telescope design based on the SC optics. A 9◦ FoV can then
be realised with a camera with ∼2000 pixels and with a diameter of
only ∼35 cm.6 The use of an SC design requires a focal plane with a
curved surface in contrast to conventional one-mirror designs which use
cameras with a plane focal plane.
Cherenkov light from EASs peaks at a wavelength of ∼350 nm at
ground level with a flash duration of typically only a few to a few
tens of nanoseconds. The use of fast and blue-sensitive photosensors
and high-speed digitising electronics is therefore required. To make
maximal use of the information contained in the time evolution of the
Cherenkov signal, full waveform digitisation and readout is desirable
per camera pixel. Given the expected range in impact distance from
the telescope, EASs from primary particles with energies of 1–300 TeV
produce Cherenkov images ranging in amplitude from around 250 to
many thousands of photons between 300 and 500 nm. With an effective
quantum efficiency of the MAPMs in the given wavelength range
(∼20%), this results in a dynamic range from around 50 photoelectrons
(p.e.) up to a few thousands of p.e. requiring each camera pixel to cover
a dynamic range of about three orders of magnitude.
The -ray flux from typical astrophysical sources and the cosmic-ray
background both exhibit power-law spectra, falling off with increasing
energy and resulting in an event rate above 1 TeV that implies a
maximum mean trigger and readout rate per SST of 600 Hz. However,
the night sky background (NSB) contributes with a Poisson-distributed
incoming stream of single photons independently to each camera pixel
at an expected rate of tens to hundreds of MHz. A fast topological
trigger is therefore needed to efficiently record Cherenkov images whilst
rejecting signals due to NSB photons.
An intelligent safety system should ensure the camera being operated
within defined limits (e.g. by measuring temperature, humidity, current
3. Technical design
CHEC-M contains 2048 pixels instrumented as 32 Hamamatsu
H10966B MAPMs each comprising 64 pixels of ∼6 × 6 mm2 and
arranged in the curved focal plane to approximate the required radius of
curvature of 1 m. The camera architecture is shown in Fig. 1. Front-end
electronics (FEE) modules connect to each photosensor providing fullwaveform digitisation for every channel and the first level of camera
trigger. A backplane forms a (second-level) camera trigger decision
based on the trigger signals from all FEE modules. Data is read out
from all FEE modules to data-acquisition (DACQ) boards routed offcamera via four 1 Gbps fibre-optic links. A safety board intelligently
controls power to camera components based on monitored environmental conditions whilst LED flashers located in each corner of the focal
plane provide a calibration source. An internal network switch provides
control connections to the safety board, LED flashers, and DACQ boards.
A camera server PC off-camera runs software to collect data, control, and
monitor the camera.
3.1. Camera mechanics and thermal control
3.1.1. Camera mechanics
The mechanical structure of CHEC-M is manufactured entirely from
aluminium and consists of an external enclosure with focal-plane plate,
an internal rack, a thermal exchange unit, and a manual lid assembly.
Fig. 2 shows an annotated view of CHEC-M with the major mechanical
elements highlighted.
The focal-plane plate located at the front of the camera is responsible
for the accurate positioning of the photodetectors. As shown in Fig. 3,
the FEE modules are slotted through this plate and into the internal rack.
On the rear of the rack is an aluminium plate with through-holes and
The result of rough calculations is rather ∼30 cm but more detailed
calculations and simulations, taking into account the final optical specifications
of the telescope (focal length of 2.283 m) and a correction due to distortion,
results in a camera diameter of 36.2 cm.
J. Zorn et al.
Nuclear Inst. and Methods in Physics Research, A 904 (2018) 44–63
0.035 mm for an ideal set of MAPMs. In reality the measured range in
the depth of the 32 MAPMs purchased for CHEC-M is ±0.26 mm. The
tiling of the curved focal plane with flat MAPMs creates an additional
maximum shift of 0.45 mm in Z from the ideal position for the edge
pixels. Combining these tolerances implies a less than 10% degradation
in telescope point spread function during operation.
3.1.2. Thermal control
The total power dissipation of CHEC-M during normal operation is
∼450 W. The thermal control system is designed to keep the camera
temperature stable over a wide range of ambient temperatures up to the
maximum required 25 ◦ C during normal operation and to protect the
camera electronics up to an ambient temperature of 45 ◦ C. A breather
desiccator removes humidity from the camera interior.
The thermal control system consists of four fans coupled to a liquidcooled heat sink. The fans, together with a system of baffles, provide
a recirculating airflow within the sealed camera enclosure. A commercially available chilling unit (Rittal SK 3336.209) provides a flowing
liquid (R134a: water glycol mixture) of controllable temperature, delivers a cooling power of ∼1.5 kW, and can operate over an ambient
temperature range of −20 to 45 ◦ C. The unit weighs 97 kg (without
liquid) and measures 485 mm × 965 mm × 650 mm. It is installed at
the azimuth axis of the telescope and is connected by 3/4’’ pipes to
the thermal exchange unit inside the camera. Quick-release non-leak
couplings allow the chiller and the camera to be disconnected quickly
from the telescope structure whilst preventing fluid loss/spillage.
Fig. 2. The CHEC-M prototype camera, with major elements indicated.
3.2. Photosensors
The Hamamatsu H10966B MAPM has a super-bialkali photocathode
with a spectral response between 300 and 650 nm, peaking at ∼340 nm
with a quantum efficiency of ∼30%. An MAPM consists of 64 pixels of
size 6 mm × 6 mm corresponding to an average angular size of 0.15◦
when installed on the GCT telescope structure. The value of 80 (the
diameter which contains 80% of the light resulting from a point source)
for the telescope design is smaller than 6 mm over the full camera FoV
once the telescope mirrors have been aligned. Due to the arrangement
in a curved focal plane, a gap of ∼2 mm between the front of the
MAPMs is required to accommodate their depth of 25.8 ± 0.26 mm.
When combined with the dead space at the edges of each MAPM, a total
maximum dead space of ∼5 mm (corresponding to the gap between the
corners of two MAPMs) is achieved. Each MAPM accepts a single high
voltage (HV) source of 800–1100 V (generated on the TARGET modules
and routed to the corresponding MAPM by an insulated cable).7 The
gain ranges from 4 × 104 to 6 × 105 , adjustable by setting the HV.
Due to the background rate associated with exposure to the night sky
(expected to lie between 15 and 500 MHz at the CTA site depending on
the observation conditions8 ) the gain for nominal operation is 8 × 104 .
The 10%–90% risetime of the MAPM is ∼0.4 ns while the transit time
is ∼4 ns with a spread of 5%–10%. Thus, the MAPM produces pulses
with an FWHM of about 1 ns. The response of the MAPM extends from
a single p.e./pixel to thousands of p.e./pixel (cf. Section 5.7).
Fig. 3. A photograph of the CHEC-M camera without the external enclosure in
place, taken from the back to highlight the backplane. The TARGET modules can
be seen inserted into the internal rack mechanics, whilst the DACQ boards can
be seen at the top of the rack, attached to the backplane via two large Samtec
ribbon cables.
screws for securing the FEE modules and for electrical connectors. Once
all FEE modules are secured, the photodetector units are attached. The
interface backplate at the rear of the camera provides a stable mounting
point for attachment to the telescope structure.
There are two access panels in the enclosure sides, into which
machined removable aluminium panels are fastened. One access panel
contains the feed-through for power cables and optical fibres via bulkhead connectors, whilst the other houses the thermal-control assembly.
CHEC-M includes a manually operated prototype lid to protect against
dust and liquid ingress. A wind shield on one side of the enclosure
provides protection for the lid whilst open. The camera is painted with
corrosion-resistant automotive paint.
Overall rigidity holds the focal plane position stable to ±0.2 mm.
Measurements on the camera mechanics indicate that the centre of all
MAPMs may be placed to within 0.35 mm of the ideal position in the
direction of the optical axis (). This misalignment may be compensated
for by a translation and rotation at the rear of the camera when installed
on-telescope using an adjustable camera mounting mechanism and
actuators on the telescope secondary reflector. This would reduce the
spread between the centres of the MAPMs in the  direction down to
3.3. Front-end electronics
The FEE module developed for CHEC-M, and shown in Fig. 4, consists
of a preamplifier module connected to a TARGET module based around
four TARGET 5 ASICs, 16-channel devices combining digitisation and
trigger functionalities [24].
The MAPM operates with negative HV, i.e. the anode is held at negative
potential while the cathode is held at ground. However, for simplicity, positive
HV values are used throughout the paper.
Throughout this paper, the NSB rate is defined as number of p.e. induced
by NSB photons per pixel and per second.
J. Zorn et al.
Nuclear Inst. and Methods in Physics Research, A 904 (2018) 44–63
Fig. 4. An MAPM attached to a CHEC-M FEE module consisting of preamplifier–amplifier module, ribbon cables and TARGET module based around four TARGET
5 ASICs.
Source: Reproduced from [23].
nominally set to 96 ns for CHEC-M (chosen to capture high-energy, offaxis events and/or events with a high impact parameter as they transit
through the FoV).
A Xilinx Spartan-6 field programmable gate array (FPGA) on board
each TARGET module is used to configure the ASICs and other module
components, to read out raw data from the ASICs, and to package and
buffer raw data for output from the module. Module control and raw
data output are managed via user datagram protocol (UDP) over a
1 Gbps Ethernet link at the rear of the modules. The TARGET 5 ASIC
is continuously sampling and dead time free, i.e. sampling continues
while data is being digitised.
The TARGET 5 ASICs also provide the first level of triggering for the
camera. The trigger consists of the analogue sum of a square of four
neighbouring pixels (referred to herein as a superpixel), which is then
discriminated. Each ASIC outputs four digital trigger signals, which are
routed through the module to the backplane, resulting in 16 differential
LVDS trigger signals per module and 512 (32 × 16) in total for the whole
Each FEE module accepts a 12 V input for all electronics use and
consumes roughly 7–8 W of power during full operation.
The MAPMs produce narrow pulses that must be shaped to optimise
the camera trigger performance. Simulations show that the optimal
pulse FWHM for triggering is around 5–10 ns with a 10%–90% risetime
of 2–6 ns. If the pulses are faster, the time gradient of Cherenkov images
across neighbouring pixels forming the analogue sum prevents pile-up
to reach the trigger threshold. If they are slower, NSB photons limit the
performance of the camera trigger.
The preamplifier module connects directly to the photosensor to
amplify and shape the signals and to provide noise immunity for signal
transport to the TARGET module. The preamplifier circuit contains an
AD8014 operational amplifier operated in trans-impedance mode and
consumes ∼1 mA quiescent current. The overall power consumption of
the preamplifier lies between 9 and 20 mW per channel depending on
the incoming photon rate. Individually shielded ribbon cables minimise
the influence of noise and provide the connection to the curved focal
plane, allowing the use of a planar internal rack to house the modules.
Each TARGET module provides 64 channels of digitisation and firstlevel triggering. For this purpose, both the signal and a reference signal
(an input pedestal voltage, Vped, supplied by an external digital-toanalogue converter (DAC), and used for common-mode noise rejection
and as a reference to fix the trigger threshold) are input to each ASIC
and simultaneously processed for sampling (data path) and triggering
(trigger path). The TARGET 5 ASIC is an analogue sampling chip capable
of digitising signals with 12-bit resolution. When used within CHECM, it provides an effective dynamic range of 1 to ∼500 p.e./pixel
(with the recovery of larger signals offline possible due to the fullwaveform digitisation). The sampling rate is tunable, but is set to
1 GSa/s for CHEC-M. TARGET 5 contains two capacitor arrays – a
64 ns deep analogue sampling array followed by a storage array with
a maximum depth of 16 384 ns – to simultaneously achieve a large
analogue bandwidth and a deep buffer. Acquisition occurs in one group
of 32 cells in the sampling array while the charge of the cells of the
other group is transferred to the storage array cells. Such a ping-pong
approach provides continuous sampling (cf. [21] for further details).
The position of the readout window digitised from storage array is
selectable with 8 ns resolution9 with a size settable in 32 ns blocks,
3.4. Back-end electronics
The back-end electronics (BEE) for CHEC-M consist of a backplane
and two DACQ boards.
3.4.1. Backplane
The backplane provides the power, clock, trigger, and data interface
to the FEE modules. Data links to the FEE modules are routed via the
backplane to DACQ boards.
The backplane triggering scheme is implemented in a single Xilinx
Virtex-6 FPGA, referred to herein as the trigger FPGA (TFPGA). The
TFPGA accepts all 512 first-level trigger lines from the FEE modules
(their signal width currently being set to 30 ns) and implements a
camera-level trigger algorithm (currently requiring a coincidence between two neighbouring superpixels). Following a successful camera
trigger, a readout request consisting of a serial message with a 64-bit
nanosecond counter (known as a TACK message) is sent to the FEE
modules to initiate a full camera readout. On the FEE modules the
TACK is compared to a local counter to determine a look-back time
in the ASIC buffers. The TACK is added to the raw data event on
This means that the start of the readout window occurs on an 8 ns edge
with respect to the nanosecond accurate event trigger (known as TACK, see
Section 3.4). This start position is identical for all pixels in a given event. The
TACK is used offline to correct the waveforms (shift them by up to 8 ns) such
that a recovered pulse can always be found at the ‘‘correct’’ position.
J. Zorn et al.
Nuclear Inst. and Methods in Physics Research, A 904 (2018) 44–63
using a UDP server which runs on the microprocessor. The SPI control
of the backplane is provided through another UDP server running only
on one of the two DACQ boards. The boards are custom-made revisions
of network switches based on White Rabbit technology [25] which are
also commercially available from the company Seven Solutions.
each FEE module and – as already mentioned in Section 3.3 – is later
used to shift the waveform and thus correct for the 8 ns resolution
of the starting position of the readout window. The TFPGA provides
functionality to individually disable any of the 512 FEE module trigger
inputs from the camera trigger decision to prevent noisy trigger patches
from dominating the event rate. The 512-bit pattern causing the last
camera trigger is accessible, and 512 counters provide a method to
monitor trigger rates across the camera.
Rate control is implemented via a settable minimal time between
consecutive camera triggers (trigger hold-off time). During commissioning it was observed that the TARGET 5 ASIC produces noise whilst
digitising previously sampled and stored analogue data. This noise
corrupts any new data sampled and stored in the ASIC whilst this
digitisation is ongoing. This is why in CHEC-M digitising is only enabled
when a readout is requested and all data sampled within the time of
the digitising process (∼20 μs) has to be discarded from the analysis.
Furthermore, it was observed that triggering with the TARGET 5 ASICs
based on signal discrimination leads to additional, false triggers due to
pick-up by the trigger circuitry of the serial data signals from the FPGA
used to read out the ASICs. Thus, a trigger hold-off time between triggers
of 80 μs (corresponding to slightly longer than that required to readout
the TARGET 5 ASICs) is used in CHEC-M to allow stable operation.
Whilst both of these problems (noise due to digitising and readout) are
solved by design in future FEE iterations (see Section 7), they lead to an
overall dead time of CHEC-M of 80 μs in nominal operation.
A second, smaller house-keeping FPGA (HKFPGA), Actel A3P400,
provides access to status and monitoring registers on the TFPGA and
monitors the current and voltage supplied to the FEE modules. Control
and monitoring of both FPGAs is provided via a serial peripheral
interface (SPI) link routed to one of the DACQ boards.
Clocks between the backplane and the FEE modules are kept in-sync
through a low-skew fan-out network and signals between the backplane
and the FEE modules are used to synchronise local time counters.
A reference clock with a frequency of 62.5 MHz is provided to the
backplane from a DACQ board. The backplane utilises a programmable
quad clock generator to generate a 125 MHz clock from the reference
clock for distribution to the TFPGA, HKFPGA, and the FEE modules.
Each of the four outputs is phase programmable in 20 ps increments and
is routed to a 1:16 fanout buffer, specified to introduce no more than
25 ps delay between the outputs. While absolute time synchronisation
will be present in the final CHEC design (see Section 7), it is not present
in CHEC-M.
The camera may be externally triggered via an external pulses input
to an SMA connector on the backplane, and routed to a bulk-head
connector on the camera enclosure. For power, the backplane accepts
a single 12 V input and generates all required voltages on a daughter
board mounted perpendicular to the main printed circuit board (PCB).
3.5. LED calibration flashers
CHEC-M is equipped with four flasher units, each containing ten
LEDs of different brightnesses placed in the corners of the focal plane
to illuminate the photodetectors via reflection from the telescope secondary mirror (cf. [26]). A Thorlabs ED1-C20 one-inch circle-pattern
engineered diffuser is mounted in front of the flashing LEDs. The
LED flasher units are based around fast gated TTL drive pulses and
3 mm, low self-capacitance, Bivar UV3TZ-400-15 LEDs, with a peak
wavelength of 400 nm. The Bivar LEDs are enabled/disabled by an
on-board microcontroller and triggered via an external TTL pulse. An
LED controller based on an Arduino Leonardo ETH board connects all
flasher units and provides an interface to set the LED pattern and to
fan out a trigger signal. The TTL trigger signal is in-turn input to the
LED controller from an SMA connector mounted on the camera chassis.
While in CHEC-M, the flasher units can only be triggered from an
external source, this will be different in the final CHEC design where
triggers can also be provided by an internal device (cf. Section 7).
Communication with the LED controller is via a network switch installed
on the internal camera rack and based on the same UDP scheme as used
for the communication with the TARGET modules and for several other
devices in CHEC-M (cf. Section 4.1).
The LED flasher units are designed to flat-field the camera across a
wide dynamic range, providing optical pulses of width ∼4.5 ns (FWHM)
at 400 nm from 0.1 p.e./pixel, for absolute single-p.e. calibration
measurements, to over 1000 p.e./pixel, to characterise the camera up
to and at saturation (cf. Section 5.9 for results of characterisation
measurement). Since the time distribution of the flasher signals in the
camera pixels can be calibrated, absolute single p.e. calibration using
the flasher units is expected to be possible even under the presence
of NSB with an expected nominal rate of about 15–25 MHz on the
CTA site. In addition, one of the SST prototypes features a shelter
to protect the telescope and camera from the environment. In such a
case, flasher calibration measurements would be performed with closed
shelter, i.e. without the presence of NSB.
3.6. Safety, power, and control
3.6.1. Safety system
The camera safety system provides the capability to remotely control
power to camera components, control and monitor fan speeds, monitor
component supply voltage and current draw as well as internal camera
temperature and humidity. Furthermore, it monitors the status of the
camera subsystems to prevent or reject actions taken by the user that
would endanger the camera, issues alerts when certain conditions are
met (e.g. temperature limits exceeded or communication lost), and
automatically takes actions to minimise the risk of damage if the
situation persists (e.g. switch off camera), i.e. if the user or software
has not taken any action first to change the situation within a defined
time window (alert-action-feedback).
The safety system consists of a power board and safety board
mounted internally in the camera on the side of the FEE rack. The power
board distributes 12 V to camera components via relays controlled
from the safety board and provides analogue monitoring of camera
component voltage supply and current draw to the safety board. The
safety board contains a microprocessor controlling the power board
relays, the digitisation of current and voltage readings from the power
board, the reading of sensors, and the alert-action-feedback. An external
high-current relay mounted on the internal camera chassis controls
power to the backplane and FEE. Communication with the safety board
is based on the same UDP interface as used for the LED controller.
3.4.2. DACQ boards
The DACQ boards form a link for raw data and communications
between the FEE modules and the camera server PC. Each board
connects 16 FEE modules via wired 1 Gbps Ethernet links to two 1 Gbps
fibre-optic links to the PC. Network interface cards (2 × Intel PRO/1000
Dual Port PCIe) are used on the PC to connect the fibres. Data is sent
to and from the FEE modules via a custom format over UDP. Jumbo
frames are used to minimise the number of packets sent per raw data
event. For each event, an FEE module serialises data from 64 pixels into
two UDP packets. Buffering on the FEE and controlled delays between
packet sending prevents the 1 Gbps (up-)links to the PC from being
saturated by the traffic from the 32 1 Gbps links to the FEE modules.
The DACQ boards each act as a layer-2 switch, with the MAC addresses
of eight FEE modules hard-mapped to a single 1 Gbps link to the
PC. Each board is based around a Xilinx Virtex-6 FPGA providing 18
GTX serial transceivers and an ARM Atmel microprocessor running an
entire light-weight Linux system for managing purposes. An Ethernet
connection to each DACQ board enables controlling and monitoring
J. Zorn et al.
Nuclear Inst. and Methods in Physics Research, A 904 (2018) 44–63
5. Continuously buffering and writing of events without loss is
ensured up to a rate of 600 Hz (corresponding to 2.5 Gbps).
6. The software is easy to adapt as soon as a final data format or
pipeline for CTA is in place.
3.6.2. Power supply
A single power supply from the company ARTESYN (iMP series)
provides CHEC-M with 12 V at up to 60 A for all electronics. The
power supply contains two individually controllable 12 V units. One unit
provides power to the camera safety system for fans, safety and power
board. The power board then distributes 12 V to the camera internal
network switch, DACQ boards, and LED controller via relays controlled
from the safety board. The second 12 V unit of the power supply provides
power to the backplane and FEE via a high-current relay, also controlled
from the safety board. The division of power distribution in this way
allows safety-critical systems to be controlled independently from highcurrent components.
The power supply weighs ∼1 kg and measures only 60 mm × 120 mm
× 250 mm. As such it can easily be housed at the rear of the secondary
mirror of the telescope. A ‘‘sense’’ feed-back input from the camera
ensures the desired voltage at the camera. A single Chainflex CF10 12way cable with outer diameter of 19 mm connects the power supply
to the camera and is flexible down to −35◦ C. The power supply can be
externally controlled and monitored using basic SMBus protocols built
on top of I2 C. An Ethernet-SMBus interface board has been implemented,
based on an Arduino Leonardo ETH board, to allow easy Ethernet
control using the same user-defined protocol based on UDP as for the
LED controller and safety board.
Two external libraries (CFITSIO [28] and simple network management
protocol (SNMP) [29]) and three self-developed libraries are used
within the software: TargetDriver (for control and readout of TARGET modules), TargetCalib (for applying TARGET module calibration),
and TargetIO (for reading and writing data from TARGET modules).
The software is structured so that each hardware component inside
(e.g. TARGET modules) and outside (e.g. chiller) the camera is represented by its own lower level class. The main class CameraInterface
serves as interface to all lower level classes. Fig. 5 illustrates the software
architecture with dependencies and libraries used. The user/client,
which can be represented by a C++ executable, a Python script, or
a GUI, can connect to the Camera Server (running on the camera
server PC physically connected to the camera) via Ethernet. The server
is implemented as a state machine, linking the client commands to
functions and state transitions defined in the main interface class. The
states are
1. Off
2. Safe
3. StandBy
4. Ready
5. Observing
6. Engineering
and serve as preliminary placeholders for final camera states to be
defined by CTA. Several safety features are implemented in the software,
e.g. timeouts, guaranteeing in addition to the safety system that the
camera is operated in safe conditions.
The hardware components are configured through ASCII files following a custom, but simple, format. Communication with all hardware
components except the chiller (which uses SNMP) is via a simple custom
protocol based on UDP (originally designed for communication with the
TARGET modules).
The camera readout and event building based on the 2048 full
waveforms provided by the TARGET ASICs of the 32 TARGET modules
is implemented in and managed by the TargetDriver and TargetIO
libraries. Events arrive at the camera server PC in asynchronous sets of
64 UDP packets (2 packets per event per module). They are first buffered
and then assembled into associated events based on the TACK in each
packet header which serves as a unique event identifier. A timeout
is used to prevent the PC buffer filling up if (in an unexpected case)
events with missing packets arrive or if the event building takes longer
than expected. Missing packets are not requested again and incomplete
events are discarded. In a subsequent step, the events are written to disk
as FITS files [28]. Once CTA is operating and a final CTA data framework
exists, the data will instead be further processed in a pipeline. However,
even then it is planned to continue developing and using TargetDriver
and TargetIO for single test purposes.
Due to the modest event rate per SST in CTA (600 Hz, cf. Section 2),
no inter-telescope hardware array trigger is required. When a telescope
triggers, all data is read out and transferred from the camera to be
processed by the software array trigger system. Decisions on whether
to proceed with the ‘‘array event building’’ in software including other
camera events from neighbouring telescopes will then be based on the
different camera event timestamps.
4. Operation procedures
In this section, the camera operation procedures are outlined. Operation procedures cover the standard control of the camera, the data
acquisition procedure, and the waveform processing methodology. In
general, there are two different types of measurements which are
∙ EAS data taking and tests being executed every time the camera
is booted (like specific software and hardware tests as well as
pedestal and transfer function measurements, cf. Section 4.2) and
∙ specific measurements performed while commissioning like flat
fielding measurements, trigger threshold and hardware temperature dependence determination. Such specific commissioning
measurements and their results are described in detail in Section 5.
As mentioned previously in Section 3.2, an MAPM gain of 8 × 104 ,
corresponding to an HV of ∼800 V, is expected to be used for nominal
data runs at the CTA site. However, due to the intrinsic performance of
MAPMs, single photoelectron (SPE) can only be resolved when operating
at a gain higher than nominal, corresponding to 1100 V. Hence, most
of the characterisation and performance tests (described in Section 5)
were done at this HV.10 The effect of lowering the gain is discussed in
Section 5 for each performance parameter separately if relevant.
4.1. Control software and data acquisition
The camera control and readout are managed by a C++ software
package referred to as CHECInterface. It is designed to be maintainable,
simple, and robust, and to fulfil the following requirements:
1. The number of dependencies on external software is at a minimum.
2. Only minimal code changes are necessary when a camera hardware component is upgraded.
3. The code is written in C++.
4. Scripts and programs written in other programming languages
are only allowed for user interface, test programs, and other
executables, but not for core-functionality.
4.2. Data calibration and waveform processing
The 2048 raw data waveforms are calibrated and processed in different steps. These consist of (1) applying ASIC specific calibration, (2)
signal charge extraction and conversion to p.e., and (3) image cleaning
and data reduction. Fig. 6 shows camera images and waveforms for the
different steps in the data calibration and processing and impressively
illustrates the need for calibration.
(1) Since the response of each of the 16 384 ASIC storage array cells
(one storage array per pixel, cf. Section 3.3) to the reference voltage
Such an approach of increasing the HV, especially for SPE measurements, is
a common procedure also used by operating IACTs like VERITAS, cf. [27].
J. Zorn et al.
Nuclear Inst. and Methods in Physics Research, A 904 (2018) 44–63
Fig. 5. Diagram showing different classes, their dependencies, and network protocols used in the camera control software CHECInterface.
Fig. 6. Camera image and waveform for different steps in the data calibration and processing for the same Cherenkov event. Uncalibrated camera image at  = 48 ns
(top left) and uncalibrated (raw data) waveform (top right) of pixel 1162. Calibrated camera image after charge extraction via the ‘‘neighbour peak finding’’ method
(bottom left) and calibrated waveform (bottom right) with pedestal subtraction, transfer function correction, and signal-to-p.e. conversion applied for pixel 1162.
The procedure of calibration and charge extraction is explained in the text. To get the y-unit of p.e./ns, the samples (in units of V) are divided by the SPE value (in
units of V ns/p.e.), determined in single-p.e. measurements (cf. Section 5.3). In both camera images, white squares indicate pixels which survive image cleaning (see
text for reference).
Vped (cf. Section 3.3) is different, the pedestal of each cell has to be
measured and then subtracted from the raw data. A run with 20 000
externally triggered events provides enough hits per cell to calculate the
mean pedestal of each storage array cell. In addition to that, the response
of each of the 64 ASIC sampling array cells (one sampling array per pixel,
cf. Section 3.3) on the supplied voltage is different. Thus, depending on
which sampling array cell is hit, another conversion between measured
ADC and signal voltage has to be used. This conversion is measured
by recording the sampling array cell pedestal (in ADC) as function of
the supplied cell voltage (given by Vped). A run with 1000 externally
triggered events per Vped provides enough hits per cell to calculate
its mean. Such a measurement results in the so-called transfer function
(see Fig. 7). Its slope depends on the ASIC parameter Isel [24] which is
adjusted to maximise the range in which the transfer function is linear.
Once determined, the values of Vped and Isel are set to default values
and do not need to be adjusted for different measurement purposes.
(2) After subtracting the pedestal and correcting for the transfer
functions, the samples of the waveform are converted into units of
J. Zorn et al.
Nuclear Inst. and Methods in Physics Research, A 904 (2018) 44–63
Fig. 8. TACK Efficiency (number of readout requests divided by number of
external triggers) and data packet efficiency (number of data packets divided
by product of data packets per event (64) and number of external triggers) as a
function of mean random trigger rate for two different backplane hold-off times
(written in brackets in the legend).
Fig. 7. Transfer function (conversion between ADC value and amplitude in
volts). The range of all ASIC sampling cells of all pixels and an example of a
single cell are shown.
p.e./ns. In order to perform this conversion the SPE value (in units
of V ns/p.e.) for each pixel is used. This value is determined from the
analysis of single-p.e. measurements, which were processed using the
same procedures as described in this section, but with this conversion
omitted. The SPE measurements are described in Section 5.3. For onsite calibration, LED calibration flasher runs described in Section 5.9
will instead be used to obtain the SPE value per pixel. Following that,
the signal charge per pixel is extracted by integrating the pulse signal.
Several algorithms to find the pulse in the waveform are currently under
investigation. One of them is the ‘‘neighbour peak finding’’ algorithm
which averages the waveforms in the pixels neighbouring the pixel of
interest and then takes the time of the maximum value in that averaged
waveform as peak time for the pixel of interest. Then the signal is
integrated in a defined window around the peak. The default size of
the window is 7 ns, with a shift to the left of the peak time of 3 ns.
A correction is then applied that uses a single reference pulse shape
for the camera to determine the percentage of the pulse outside of the
integration window. In doing so, the method is less dependent on the
integration window size used for the determination of the number of
photoelectrons inside the pulse, however all the analysis performed for
this paper did keep the same window size of 7 ns.
(3) Similarly to the charge extraction methods, several algorithms for
image cleaning and data reduction are under investigation. A possible
procedure is the tail-cut approach, where all pixels containing a signal
greater than a threshold are retained, and all other pixels above a second
lower threshold are retained if they are a neighbour to a pixel that
satisfies the first criterion.
The results of calibration and waveform processing are waveforms
and camera images as shown in the lower panels of Fig. 6. The software
being used in (2) and (3) is the low-level data processing pipeline
software ‘‘ctapipe’’ [30], currently under development for CTA.
is used.11 Its intensity can be regulated using a filter wheel settable to
attenuations in the range of 1 to 5 × 104 and diffusers are used to obtain
a uniform illumination of the camera focal surface. The experimental
set-up of the light source is similar to the one presented in [31]. The
maximum spread of the illumination across the camera focal surface
is measured (using an SiPM with known gain and temperature-gain
dependence and a robot arm to scan the laser beam) to be ∼1%.
5.1. Data rate
The ability of the camera to read out waveforms from all 2048 pixels
and send them in UDP packets to the camera server as function of the
trigger rate is assessed in this section.
To measure the data packet efficiency (number of arrived data
packets divided by number of expected data packets12 ) as a function of
the trigger rate, the camera was externally triggered by pulses randomly
distributed in time. Cases with two different backplane hold-off times
(cf. Section 3.4.1) of 80 μs and 200 ns were investigated. Since the
camera is triggered externally, no additional triggers due to sampling
are expected in the latter one. Fig. 8 shows the result of such rate
measurements. It can be observed that
∙ in case of no (or a very low) artificial backplane hold-off time
(of 200 ns), an efficiency of 1 (±0.1%) is observed for camera
trigger rates smaller than ∼900 Hz. The uncertainty of 0.1% is
due to the (in)accuracy of counting the pulse generator triggers
in a given time window,
∙ in case of an artificial backplane hold-off time of 80 μs, there is
a chance of two or more (random) triggers arriving within that
time. This chance increases with increasing mean trigger rate.
Note that for all laser illumination measurements presented in this paper,
the deliberate shaping of electrical pulses by the camera preamplifier circuit
dominates the intrinsic width of the laser pulse so that the FWHM of the pulses
measured by the camera is quite stable as long as no saturation effects occur (cf.
Section 5.5).
The number of expected data packets is the product of the number of
external triggers and the number of packets per event being 64 (2 packets for
each module).
5. Characterisation and testing
In this section, the camera characterisation based on results of
camera lab tests during commissioning and their impacts on the camera
performance and characterisation are presented. For illumination measurements, a laser of wavelength 398 nm with a pulse FWHM of 80 ps
J. Zorn et al.
Nuclear Inst. and Methods in Physics Research, A 904 (2018) 44–63
spread can be reduced by ‘‘gain matching’’ the camera, i.e. by supplying
each MAPM with a different HV. Three mean HV values (800 V, 900 V,
and 1000 V), each of them consisting of 32 different HVs providing
the same mean gain for all MAPMs, were determined in specific laser
measurements where the laser amplitude was kept fixed at a medium
illumination level while the HV was altered. The gain spread over the
camera after gain matching is reduced, but still around 30% (see green
distributions in Fig. 9b). This is due to the fact that the HV can only be set
individually for each MAPM, not for each pixel (a fundamental feature of
the MAPM design). However, since the camera was illuminated with the
uniform light source, the remaining gain differences between the pixels
could be used to define pixel dependent flat field coefficients, which are
included in the SPE conversion factors described in Section 4.2.
This is why the packet efficiency is less than 1 at rates below
900 Hz and decreases consistently with the fraction of readout
requests, and
∙ the packet efficiency data points of both cases lie above each
other at rates higher than ∼900 Hz. At this rate, the efficiency
decreases (while the readout request fraction continues with the
same slope) because the maximum transfer rate of the DACQ
boards is reached causing packet loss being independent of any
backplane hold-off time.
To conclude, due to the artificial hold-off time of 80 μs implemented
on the backplane to avoid additional triggering on the trigger circuitry,
the camera suffers a 5% data packet loss at the requested mean random
rate of 600 Hz in normal operation mode. However, it is shown here
that once this problem is solved (which is the case in the next TARGET
ASIC generation, cf. Section 7), no losses will occur at the requested rate
of 600 Hz using UDP to send the data to the camera server.
5.4. Trigger threshold determination
As described in Section 3.3, the TARGET 5 ASICs provide the firstlevel trigger of the camera by discriminating the analogue sum of
four neighbouring pixels (referred to as superpixel). The threshold for
discrimination is set individually for each superpixel by the combination
of two ASIC parameters: Pmtref4, setting the reference voltage for
the summing amplifier, and Thresh, setting the reference voltage for
the comparators. There is a certain range of combinations of those
parameters for which the trigger functions properly and different combinations can lead to the same trigger threshold, resulting in slightly
different trigger noise levels (cf. [24] for detailed information and
test results). Thus, each full camera trigger threshold setting consists
of 512 (possibly) different pairs of Pmtref4 and Thresh. To identify
these values, the camera was uniformly illuminated at different laser
amplitudes while counting (with the backplane TFPGA) the number of
triggers for each superpixel (first-level trigger) for a given light/HV level
and (Pmtref4/Thresh) pair individually. The final (Pmtref4/Thresh) pair
was chosen so that the superpixel trigger efficiency is about 50% for
the given laser amplitude in each superpixel. In this way, five different
threshold settings (each with 512 pairs of Pmtref4 and Thresh) at mean
illumination levels of around 2, 5, 11, 29, and 78 p.e./pixel at each of
the three previously defined gain-matched HV settings were defined.
Fig. 11a shows the resulting mean superpixel trigger rate as a function
of the laser illumination for the five threshold settings (at a mean HV of
800 V).
In the next step, the overall camera trigger rate (second-level
trigger) was measured as a function of the camera trigger threshold.
On telescope, such measurements are typically used to identify the
camera trigger threshold operating point for given background light
conditions. To understand the influence not only of the NSB but also of
the TARGET module sampling and data sending on the camera trigger
rate, three different scenarios were investigated in the lab (results shown
in Fig. 11b):
5.2. Transfer function
As mentioned in Section 4.2, the transfer function provides a lookup table to relate ADC counts to signal amplitude in volts. This lookup table, different for each sampling cell of each pixel, is used in the
electronic calibration of the camera raw data. Fig. 7 illustrates the range
of transfer functions of all ASIC sampling cells of all pixels in the camera
for the ASIC parameters Vped = 1050 (corresponding to 650 mV) and
Isel = 2816. The chosen Vped value minimises the spread at low signal
amplitudes while the selected Isel value ensures a high dynamic range
of ∼3800 ADC counts with a reasonable linear shape of the transfer
5.3. Single photoelectron measurement and charge extraction
Measuring the pulse area spectrum on the single p.e. level is a
fundamental step in the calibration to determine the conversion factor
between pulse area (in V ns) and p.e. (referred to as SPE value). This
factor is different for each pixel and depends on the HV the MAPM
is supplied with. Due to the intrinsic performance of the MAPM, the
most reliable SPE resolution can be obtained at the highest possible
MAPM gain, corresponding to an HV of 1100 V. This is why the SPE
measurement was done at that HV illuminating the whole camera with
a medium light level of ∼0.35 p.e./pixel. Fitting the spectrum of each
pixel with a Poisson distribution convolved with individual Gaussians
for the noise and SPE peak (example for one pixel shown in Fig. 9a) leads
to different fit parameters for all 2048 pixels such as mean illumination
level, noise peak,13 SPE value (distance between first (noise) and second
(1 p.e.) peak), and relative SPE width. The
√ latter one is proportional to
− 2
, where Noise is the
the excess noise factor and defined as
width of the noise peak, SPE the width of the SPE peak, and   the
SPE value. Distributions of fit parameters including all 2048 pixels are
shown in Fig. 9b (SPE value, rightmost blue distribution) and Fig. 10
(illumination, noise peak, and relative SPE width).
To measure SPE values at HVs less than 1100 V, in the first step
the camera was illuminated at a higher illumination of ∼100 p.e./pixel
at 1100 V. The charge in p.e. was determined by calculating the pulse
area and using the previously determined SPE value. In the second step,
the HV was reduced keeping the illumination at the same level and
thus the number of p.e. constant. However, since the gain , being the
amplification factor of each pixel, decreases with decreasing HV, the
pulse area also does so. This results in a lower SPE value, determined by
measuring the pulse area at this lower HV and using the known number
of p.e.. The blue distributions in Fig. 9b show the SPE values of all pixels
when all MAPMs are set to the same HV (indicated on the x-axis). The
1. Backplane hold-off (BO) time set to 80 μs as used in CHEC-M
for nominal operation (BP HO on) and all TARGET modules
configured to not sample and not send data (SD off) — blue data
points in Fig. 11b,
2. BP HO on and all TARGET modules configured to sample and
send data (SD on) — green data points in Fig. 11b, and
3. BP HO off and SD off, with a white LED emulating an NSB rate
of ∼50 MHz (measured with an additional SiPM with known
gain and temperature-gain dependence) and a bright laser with a
constant rate of 600 Hz (shown as grey line in Fig. 11b) emulating
Cherenkov light of about 200 p.e./pixel — black data points in
Fig. 11b.
It was observed that SD produces additional triggers causing the trigger
rate to increase by ∼4 orders of magnitude at a camera trigger threshold
of 5 p.e./pixel compared to the scenario where SD is disabled. This
implies that the trigger circuitry picks up noise not only from the FPGA
serial data signals used to read out the ASICs (already fixed by using a
The noise peak is centred around 0 since the pedestal was subtracted for each
ASIC storage cell during calibration of the data (as explained in Section 4.2).
J. Zorn et al.
Nuclear Inst. and Methods in Physics Research, A 904 (2018) 44–63
Fig. 9. (a) SPE spectrum and fit for pixel 1559 at 1100 V. (b) Distribution of SPE value across the camera for a non-gain-matched camera (all MAPMs set to the same
HV indicated on -axis, blue) and for a gain-matched camera (MAPMs set to different HVs, where mean HV is indicated on -axis, green). Lines show the median
and the interquartile ranges.
Fig. 10. Distributions of SPE fit results for all 2048 pixels: (a) mean illumination, (b) noise peak, and (c) relative SPE width.
BP HO time of 80 μs, cf. Section 5.1) but also from the sampling and data
sending process itself. Both issues will be solved by design in future FEE
iterations (see Section 7). This is why the third measurement in Fig. 11b
(black curve) shows how the rate curve is expected to look like in future
FEE iterations under the influence of NSB and Cherenkov showers. In
this example, the camera trigger rate first decreases from ∼50 MHz at 2
p.e./pixel, where the trigger is completely dominated by the emulated
NSB, down to 600 Hz at a camera trigger threshold of 29 p.e./pixel. From
this point on, the camera trigger is dominated by the laser, emulating a
Cherenkov light signal.
For an SST with a CHEC-M-like camera the NSB level on the CTA
site is expected to lie between 15 and 25 MHz (lower than in the lab
measurements shown in Fig. 11b). Thus, the rate curve is expected to
flatten at lower trigger threshold compared to Fig. 11b, which means
that a range of trigger threshold settings between 2 and 100 p.e./pixel
should be sufficient. It is useful to determine additional intermediate
trigger threshold settings. This can be done either by performing a
finer and thus more time-consuming (Pmtref4/Thresh) scan or by
interpolating the (Pmtref4/Thresh) values between two settings. The
latter method was used in two of the three trigger rate measurements
presented above (blue and green data points in Fig. 11b, intermediate
steps between the previously determined five trigger threshold settings).
This procedure for trigger threshold determination was used to
produce fine-grained steps of approximately 20 per decade in threshold
for the use of on-sky measurements. Excessively noisy trigger pixels
were disabled at each threshold setting until the second-level trigger
rate stabilised. For the next CHEC camera prototype (cf. Section 7), the
procedure will be different: The individual ASIC parameters (Pmtref4
and Thresh) will be determined during the commissioning of the new
TARGET modules by injecting electrical signals. Thus, the trigger threshold determination is disentangled from photosensor characteristics like
gain, quantum efficiency, etc. and a camera trigger threshold determination to a precision of <1 p.e./pixel will be possible.
5.5. Pulse shape characteristics
As explained in Section 3.3, to optimise the trigger performance, the
signal pulse FWHM and 10%–90% pulse risetime should lie between
5 and 10 ns and between 2 and 6 ns, respectively, over the whole SST
J. Zorn et al.
Nuclear Inst. and Methods in Physics Research, A 904 (2018) 44–63
Fig. 11. (a) Camera trigger rate as function of the trigger setting and illumination (in p.e./pixel) for a mean HV of 800 V, laser triggering at 1 kHz. The absolute
illumination levels were deduced from the reconstructed mean pixel charges. (b) Camera trigger rate as function of the trigger threshold for three different scenarios
all at a mean HV of 800 V: (1) BP HO on and SD off, (2) BP HO on and SD on, and (3) BP HO off, SD off, with white LED of 50 MHz and a laser with a constant rate
of 600 Hz and an illumination level of about 200 p.e./pixel — further explanation given in the text.
energy range. Fig. 12 illustrates the FWHM and risetime for two pixels14
as function of the illumination level for laser data taken at different laser
brightnesses all at 1100 V. It shows that the pulse shape requirements for
optimal triggering are met at low illumination levels and that the pulse
shape is stable up to an illumination of 200–400 p.e./pixel depending
on the camera pixel. At these illuminations, saturation effects can be
observed, i.e. while the pulse peak height stops to increase the FWHM
continues to increase with increasing laser amplitude. The amplitude
at which saturation occurs is different for each pixel due to different
gains at 1100 V and/or different quantum or collection efficiency. Since
no quantum or collection efficiency spread between different MAPMs
and pixels are reported by the manufacturer, the influence of the latter
aspect is expected to be small and much lower than the gain spread at
1100 V.
The influence of saturation can also be observed in Fig. 13 showing
the FWHM and 10%–90% risetime distribution for all 2048 pixels at
different illumination levels (same data set as used for the sample
pixels in Fig. 12). As explained above, due to the gain spread at
1100 V, saturation occurs at a different illumination level for each
pixel, resulting in a wide spread in FWHM, especially at the highest
illumination levels when all pixels are affected by saturation effects.
However, for all pixels, both the FWHM and 10%–90% risetime fulfil
the requirements for optimal triggering.
The intrinsic MAPM pulses are significantly shorter (FWHM of ∼1 ns,
cf. Section 3.2) than the pulses measured with the whole chain (FWHM
between 5 and 10 ns) being dominated by the preamplifier pulse shape
characteristics. Thus, the effect of the HV (i.e. also of a lower HV and
gain) on the pulse shape characteristics – measured at a given signal
amplitude in V – is expected to be negligible.15
5.6. Crosstalk
The crosstalk measurement was performed at 1100 V with an MAPM
connected to a CHEC-M preamplifier module which was in-turn probed
with an oscilloscope while only one pixel was illuminated with a laser
(all other pixels physically masked). The peak-to-peak voltage of the
average pulses from both the signal and candidate pixel were then
measured and the ratio taken as an indication of the crosstalk. The 64
pixels of a single MAPM are mapped in groups of 16 to preamplifier
boards, and then one-to-one to ASICs. Clear average pulses were seen
in all pixels connected to the same board as the signal pixel, resulting
in an average crosstalk of 4%–5% and reaching a maximum of 6% in
neighbouring pixels (see Fig. 14). No discernible pulses were measured
in pixels of the other three preamplifier boards and the values shown
for boards 0–2 in Fig. 14 represent the limit of the measurement
technique and should be taken as an indication that no significant
crosstalk has been measured. This measurement strongly indicates that
the preamplifier PCBs rather than the MAPM are the dominant source
of crosstalk in the system. A lower HV/gain is therefore not expected
to have a significant impact on the crosstalk. Furthermore, this result
is not entirely unexpected since in a compact, high-density system it is
inevitable that signals must be routed in close proximity on any given
PCB. However, for next CHEC iteration designs, the preamplifier board
routing has been optimised to minimise crosstalk.
The results of the crosstalk measurements have to be taken into
account in the uncertainty evaluation. A maximum crosstalk of 6%
between neighbouring pixels does not only degrade the charge resolution by the same amount, but also the SPE calibration, both affecting
the image reconstruction: Taking into account a possible gain spread
of about 30% between neighbouring pixels, high-gain pixels can bias
the SPE calibration of neighbouring low-gain pixels by about 8%
due to crosstalk. In addition, the crosstalk also affects other camera
performance aspects like the trigger efficiency.
These two pixels were chosen on a semi-random basis as qualitative
representatives of all camera pixels. They show neither particularly good nor
poor characteristics and are geometrically well separated (one is near the edge
of the camera and one at the centre), rather than e.g. being of the same FEE
Of course, since a higher HV/gain results in higher SPE values and saturation
effects affect the pulse shape characteristics (as shown in this section), a
lower/higher HV affects the pulse shape characteristics, if measured as a
function of illumination level, shifting the data points in Figs. 12 & 13 to
higher/lower illumination levels.
5.7. Dynamic range
The dynamic range of the signal recording chain (MAPM and FEE
module) was assessed by illuminating the entire camera with a uniform
light level ranging from below 1 p.e./pixel to several hundreds of
p.e./pixel in calibrated steps. The measurements were done supplying
all MAPMs with the maximum voltage of 1100 V to be able to resolve
SPE at low illumination. Results for two camera pixels (same as used for
J. Zorn et al.
Nuclear Inst. and Methods in Physics Research, A 904 (2018) 44–63
Fig. 12. Pulse FWHM (blue points) and 10%–90% risetime (green points) as function of measured peak height for camera pixel 1825 (left) and 1203 (right). The
peak height (in units of p.e./ns) was extracted from the data using the procedure explained in Section 4.2. The squares indicate different laser illumination levels,
their y-position the mean pulse FWHM/risetime, and the bars the standard deviation in measured FWHM/risetime and peak height at each laser brightness.
Fig. 14. Crosstalk (in %, indicated by colorbar) for different pixels of one MAPM
at 1100 V routed to different preamplifier (preamp) boards when only pixel 15
(indicated by red square) was illuminated with a laser (all other pixels were
physically masked). For more details see text.
Fig. 13. Pixel distributions (mean for each pixel) for pulse FWHM (blue) and
10%–90% risetime (green) for different illumination levels. Lines show the
median and the interquartile ranges.
∼250 p.e./pixel for camera pixel 1825 while it is not observed in pixel
1203 over the range of laser brightnesses used in these measurements.
This can again (as in Section 5.5) be explained by saturation effects
occurring at different illumination levels due to different gains at 1100 V
and due to different quantum and collection efficiencies between the
two pixels (less substantial). Furthermore, the full signal recording
chain (MAPM and FEE module) has a similar response as the MAPM
itself, showing non-linearity effects of about 20% at 1000 p.e./pixel.
Consistent results are obtained when looking at the dynamic range of
all pixels in the camera (see Fig. 15b).
As can be suggested from the waveforms of the two pixels at different
illumination levels, shown in Fig. 16, as well as from the FWHM
dependence on the illumination level (cf. Fig. 12), the pulse width
increases with illumination. Thus, the relationship between pulse width
at fixed amplitude (e.g. at 20 p.e./pixel) and input illumination level
can be used as a first attempt for recovery in saturation as shown in
Fig. 15a for pixel 1825 and Fig. 15b for all pixels.
The overall dynamic range can be shifted to higher illumination
levels by reducing the gain. Operating the camera at a mean HV of 800 V
instead of 1100 V reduces the gain by a factor of ∼6 shifting the upper
the FWHM and risetime investigation, cf. Section 5.5) and of an MAPMonly measurement for comparison are shown in Fig. 15a. For the camera
pixel data points, the charge (y-axis) was reconstructed following the
procedure explained in Section 4.2. Its value at ∼50 p.e./pixel was
used as anchor for the illumination level on the -axis to absolute
calibrate the laser. Other points on the -axis were then inferred from the
relative calibration of the filter wheel used to adjust the laser intensity,
resulting in an illumination in units of p.e./pixel. The MAPM data points
were inferred from measurements where the MAPM signal was directly
measured with an oscilloscope while the absolute calibration of the axis was determined from single-p.e. spectra fits at the lower end of the
range and using the relative calibration of the filter wheel for the rest
of the -axis range.
According to Fig. 15a, clear deviation from a linear correlation
between illumination level and reconstructed charge (using the procedure described in Section 4.2) starts to occur at an illumination of
J. Zorn et al.
Nuclear Inst. and Methods in Physics Research, A 904 (2018) 44–63
Fig. 15. (a) Reconstructed charge as function of the camera illumination for two camera pixels with the full chain of MAPM and FEE module, and for the MAPM
only. Points show the mean, bars the 25th and 75th percentile of the charge distribution for the given pixel and illumination. A first attempt for saturation recovery
using the pulse width at a fixed peak amplitude is shown for pixel 1825 (see text for details). The grey dashed line shows a 1:1 relation between the axes. (b) Pixel
distributions (mean for each pixel) for extracted charge at a given mean illumination level with and without attempts to recover from saturation at the highest
illumination levels using the pulse width. Lines show the median and the interquartile ranges.
Fig. 16. Calibrated waveforms of camera pixel 1825 (left) and 1203 (right) at different illumination levels. At an illumination of 580 p.e./pixel, the pulse in pixel
1825 shows obvious saturation effects (plateau at the top).
levels (see Fig. 17 for two sample pixels and for the distribution of all
pixels, respectively). It improves with increasing illumination due to
increasing signal-to-noise ratio and is (for most of the pixels) better than
1 ns for illumination levels >6p.e./pixel. Deterioration for illumination
levels higher than 110 p.e./pixel is observed in the mean of the allpixel distribution and for the sample pixel 1825, again due to saturation
effects occurring at different illumination levels.
The timing and time resolution could be affected by a changing
pulse shape. This could explain the degradation of the time resolution at
high illumination levels in the saturation regime. However as explained
previously in Section 5.5, the impact of a lower HV/gain at a given signal
amplitude in V is expected to be insignificant on the pulse shape, thus
the same holds for the time resolution.
end of the dynamic range to ∼6000 p.e./pixel, resulting in an overall
dynamic range of ∼4 orders of magnitude.
5.8. Timing
To investigate time differences between digitised signals of different
pixels hit by the same light flash simultaneously, the camera was
externally triggered while illuminated by a laser at 1100 V. For each
illumination level and pixel, the pulse peak time distribution was determined out of 500 events, where the individual peak time of each pixel
and event was shifted by the camera mean of the given event to overlap
different events. The time resolution, defined as standard deviation
of the peak time distribution, was measured for different illumination
J. Zorn et al.
Nuclear Inst. and Methods in Physics Research, A 904 (2018) 44–63
Fig. 17. Time resolution as function of the camera illumination (a) for camera pixels 1825 and 1203 and (b) for all pixels (distributions) with lines showing the
median and the interquartile ranges.
Fig. 18. (a) Flasher brightness, measured with an SiPM with known gain and temperature-gain dependence, and converted into an illumination level per camera
pixel as function of ambient temperature for 20 different LED patterns. Each line shows an individual polynomial fit of second order ( () =  2 +   + , a, b, c being
fit parameters) to the data of each pattern. Black (upper) points: brightness measured with a neutral density (ND) filter of ∼10% transmission in front of the SiPM;
red (lower) points: brightness measured without ND filter but additional amplifier to amplify the SiPM signal. Temperature dependence of second and third dimmest
pattern measured with both set-ups, third dimmest pattern used to scale. (b) Flasher brightness converted into an illumination level per camera pixel as function of
temperature cycling measurements for the same 20 LED combinations as used in (a). Different cycle lengths were used for measurements with and without (three
dimmest LED patterns) ND filter. Points showing the data (results of the measurements), lines showing the ‘‘corrected’’ data after applying the temperature correction
factor deduced from the polynomial fit of the data sets in (a).
gives different fit results for each combination which can be used to
correct for temperature effects in other data sets (see Fig. 18b). The
spread after applying the temperature correction is about 3%–8% for the
eight dimmest LED combinations decreasing to about 1% for all other
combinations. Whilst this resulting spread results in sufficient stability
to utilise the LED flashers over the expected operating range of CHECM, some correlation with temperature clearly remains in Fig. 18b. The
resulting residuals from a perfectly stable response are larger whilst the
temperature decreases — implying a hysteresis in the temperature response that is not considered here (either in the derivation or application
of temperature coefficients) and will be examined in the future. The
long-term stability (measured over the time-scale of several days) shows
a decrease in brightness ranging from 0.25%/h and ∼1%/h depending
5.9. LED calibration flashers
The four flasher units were tested for stability and temperature
dependence, as well as on their dynamic range. They were operated
in a temperature controlled environment while their brightness was
measured with an SiPM (with known temperature-gain dependence).
The overall dynamic range of one flasher consisting of 10 LEDs is about
four orders of magnitude with illumination levels at the camera in the
range of sub-p.e./pixel up to a few thousands of p.e./pixel. The flasher
brightness dependence on temperature is different for each combination
of the 10 LEDs (pattern). It shows an increase with temperature of
<1%/◦ C for the brightest pattern up to 5%/◦ C for the dimmest (see
Fig. 18a). Fitting this dependence with a polynomial of second order
J. Zorn et al.
Nuclear Inst. and Methods in Physics Research, A 904 (2018) 44–63
Fig. 19. Camera temperature as a function of time (a) measured with the TARGET module (TM) temperature sensors (upper data points) and the five safety board
(SB) sensors (lower data points) during four power cycles keeping the chiller temperature at a fixed value of 5 ◦ C and (b) measured with the SB sensors when the
chiller cooling temperature was changed by one degree every hour while the camera was on and data was taken.
on the LED combination, recovering completely after a power down of
one hour.
The measurements showed the flashers being appropriate devices
for regular camera calibration with the possibility of absolute gain
determination using the dimmest LEDs for SPE measurements and for
monitoring changes in dynamic range and linearity of the full signal
recording chain (MAPMs and FEE modules). Furthermore, the analytic
description of the temperature dependence can be used to correct for
changes in illumination in case temperature drifts occur.
5.10. Camera and temperature stability
Several outdoor camera power cycle and temperature stability measurements were done to test the camera reliability and its behaviour
with temperature.
Fig. 19a shows four power cycle measurements at a fixed chiller
temperature of 5 ◦ C starting with the first power cycle at sunrise on
a spring day at an ambient temperature of ∼5 ◦ C. It can be deduced that
the maximum temperature difference between the TARGET modules
in the camera at a certain time is about 6 ◦ C and that the modules
with the lowest temperature are those located at the top and bottom
of the camera. This is expected since all auxiliary boards like safety,
power, and DACQ boards are attached on the sides, while no boards are
located at the top and bottom, and the fans are installed at the bottom.
Furthermore, it was observed that the mean camera temperature varies
over 8 ◦ C over a complete day with the largest change occurring during
sunrise and sunset. This effect could be corrected for by changing the
chiller temperature accordingly to the ambient temperature. As shown
by temperature cycling tests (Fig. 19b), the camera temperature can
be controlled and maintained on a certain level by adapting the chiller
The temperature dependence of the baseline of all 2048 pixels was
investigated using the same chiller temperature cycle runs shown in
Fig. 19b, in which externally triggered 45-min runs at 3 Hz for each
chiller temperature were taken. In order to observe a relative change in
the baseline with temperature, all 45-min run data was subtracted from
a fast (600 Hz) reference pedestal run taken at the beginning of the
temperature cycle runs at a chiller temperature of 5 ◦ C. The resulting
baseline-shift temperature dependence is fitted with a linear function
( ) =   +  with temperature  and fit parameters  and  for the
camera mean and for each pixel individually. The results show that
Fig. 20. Camera pedestal mean and standard deviation (std) as function of
the temperature for the data set shown in Fig. 19b. The data is fitted with a
linear function ( ) =   +  with temperature  and resulting fit parameters
of  ∼ 0.36 mV/◦ C and  ∼ −8.30 mV for the pedestal mean and  ∼ 0.01 mV/◦ C
and  ∼ 0.26 mV for the standard deviation.
the mean camera baseline shift is about 0.36 mV/◦ C (see Fig. 20).
A spread in the temperature dependence between individual pixels is
observed (some of them even with opposed sign, see Fig. 21a) with
minimum and maximum temperature coefficient of min ∼ −0.26 mV/◦ C
and max ∼ 1.18 mV/◦ C, respectively. This spread can be attributed to
different temperature behaviours of either the ASICs or the external
DACs providing the input signal Vped to each ASIC.16 This is illustrated
by Fig. 21b where it can be observed that pixels connected to the same
ASIC and external DAC show similar baseline temperature coefficients.
In the next TARGET module generation, the offset Vped will not be supplied
by an external DAC anymore but by the ASIC itself.
J. Zorn et al.
Nuclear Inst. and Methods in Physics Research, A 904 (2018) 44–63
Fig. 21. (a) Distribution of all 2048 pixel baseline temperature coefficients (fit parameters  in linear fits, ( ) =   + ) giving a mean of ∼0.35 mV/◦ C and a
standard deviation of ∼0.22 mV/◦ C. (b) Camera image illustrating the 2048 pixel baseline temperature coefficients. Each block of 16 pixels is connected to the same
ASIC and external DAC on the given TARGET module.
However, even though the baseline of different pixels have different
temperature dependencies, only a small temperature dependence of
the pixel baseline standard deviation is observed (see Fig. 20). This
means that all 16 384 storage cells of one pixel have similar temperature
dependencies. Thus, one linear correction factor per pixel is enough to
characterise or correct for the temperature dependence.
An uncertainty on the baseline of ±0.5 mV is not expected to affect
the camera performance in terms of charge resolution. However, in case
the camera is not kept at a constant temperature level within ±1 ◦ C,
either a pixel dependent linear correction factor has to be used or
the pedestal must be remeasured every time the camera temperature
changes by more than 1 ◦ C. A pedestal measurement takes about 30 s
and the HV needs to be off or the lid closed. Thus, this approach would
cause a maximum dead time of ∼0.8% per night (according to a very
simplified calculation, assuming a linear temperature drift of 8 ◦ C within
an eight-hours night). To avoid that, a pedestal determination could be
done ‘‘online’’ during observation instead, either with interleaved events
or by using parts of the waveforms/pixels with no signals. The latter
option is the long-term plan for next camera iterations.
To investigate the camera warm-up, two relevant quantities were
investigated: first, the change in trigger rate with closed lid and HV
off, indicating whether the trigger threshold is stable and the camera
is ready for triggering, and second, a possible shift in the baseline
indicating whether the camera is ready for data taking. To measure
the trigger rate change during warm-up, the trigger threshold was set
in the electronic noise causing the trigger rate to be very sensitive to
electronic noise changes expected to occur during warm-up.17 Fig. 22a
shows the trigger rate change as function of the time after camera power
up. It is fitted with an exponential function () =  (1 − exp(−∕t )) + 
with time  and fit parameters , , and t , showing an increase over
time with a time constant of t ≈ 1193s. This increase is connected
with a temperature increase during warm-up (as shown by Fig. 22b)
and very likely caused by the baseline drift with temperature explained
above. An increasing baseline in fact reduces the threshold, causing a
higher trigger rate. Thus, the mean camera baseline is expected to show
an increase during warm-up which is shown by Fig. 22c. The resulting
time constant is b ∼ 849s with a mean baseline drift of about 0.35 mV
over ∼40 min being less than 1 p.e. In total, the camera is assumed
to be stabilised/warmed up when the trigger rate change or the mean
camera baseline drift is less than 5% compared to the asymptotic value
lim→∞ (). This is the case after ∼3  (∼1 h for the trigger rate change
and ∼45 min for the mean camera baseline drift).
The camera parameters investigated in this section (temperature,
trigger rate change, and baseline drift) are the most basic ones which
allow one to characterise the general functionality of the camera and to
answer the most basic questions like: Can the temperature be controlled
and stabilised? By how much does the temperature change over a
day? What is the temperature distribution and its spread inside of
the camera? Can the camera be triggered reliably? Can data be taken
reliably and is it affected by temperature or time? Other measurements
for more detailed stability and temperature analyses (like stability
and temperature dependence of timing and gain) are planned for the
next camera prototype, especially since due to the use of SiPMs the
performance of CHEC-S is expected to show more serious temperature
dependencies than the performance of CHEC-M (see Section 7).
6. Cherenkov events
First Cherenkov light with CHEC-M was observed in November 2015
during a first campaign with the camera deployed on the GCT telescope
prototype located at the Paris Observatory in Meudon near Paris [32].
A second campaign was carried out in Spring 2017.
Due to the high NSB light level in Meudon, estimated to be 20–100
times brighter than at the CTA site, the camera was operated at a low
gain (mean HV of 800 V) and at trigger threshold setting 3 (ref. Fig. 11a),
corresponding to roughly 11 p.e., pushing the trigger rate down to only
∼0.1 Hz. Two examples of on-sky events (telescope pointing to the sky,
camera lid open, HV on) and one event with telescope in park position
(0◦ elevation), camera lid closed, but HV switched on, all three recorded
during the second campaign are shown in Fig. 23. The upper images
show the intensity in p.e. (integrated charge) for each pixel, while the
lower ones indicate the peak arrival time (after trigger) for each pixel for
the same events. Furthermore, the white boxes indicate pixels surviving
image cleaning (cf. Section 4.2). As expected for a Cherenkov flash from
a shower, the timing plots of the Cherenkov events (Fig. 23d & e) show
the image propagating across the focal plane in time. Whereas the first
event (Fig. 23a & d) could have been a shower with a large impact
distance thus showing a rather large time gradient of about 35 ns, the
second event (Fig. 23b & e) could have been an inclined shower with
the telescope being at the edge of the Cherenkov light pool causing all
In observing mode, the trigger threshold will be set well above the electronic
noise level so the trigger rate will be less sensitive to electronic noise changes.
It will be dominated by NSB and Cherenkov events.
J. Zorn et al.
Nuclear Inst. and Methods in Physics Research, A 904 (2018) 44–63
Fig. 22. (a) Trigger rate change  as function of time after camera booting, fitted with an exponential function () =  (1 − exp(−∕t )) +  with time  and resulting
fit parameters of  ∼ 189 Hz,  ∼ 875 Hz, and t ∼ 1193 s. The trigger rate change at time  is defined as  =  − lim→∞ () with  being the measured rate at
time . (b) Trigger rate as function of temperature for the same data as shown in (a), fitted with a linear function ( ) =   +  with temperature  and resulting fit
parameters of  ∼ 67.74 Hz/◦ C and  ∼ −753.19 Hz. (c) Camera mean baseline drift  as function of time after camera booting, fitted with an exponential function
() =  (1 − exp(−∕b )) +  with time  and resulting fit parameters of  ∼ 0.37 mV,  ∼ 0.61 mV, and b ∼ 849 s. The camera mean baseline drift at time  is defined
as  =  − lim→∞ () with  being the measured camera mean baseline at time .
Fig. 23. Camera images of three different events, showing the intensity (a, b, and c) and the peak arrival time (d, e, and f) for each pixel. The white squares in (a),
(b), and (c) indicate the pixels that remain after the tail-cut cleaning. All modules were active in this observation run. For further explanation refer to the text.
pixels with Cherenkov signal being illuminated at a very similar time.
The event shown by Fig. 23c and f was recorded with closed lid in park
position. It must have been a cosmic ray induced particle travelling
through the curved MAPM array of the camera. The unique geometry
and fast time profile of such an event make it easy to be isolated from
Cherenkov events in the analysis afterwards.
The additional timing information in both the Cherenkov and direct
cosmic ray events is only possible due to the waveform sampling nature
of the camera electronics and is useful for advanced image cleaning,
background rejection, and event reconstruction algorithms. Additionally, images at the highest energies can take many tens of nanoseconds to
cross the camera, as can be seen in Fig. 23d. Without a ∼100 ns readout
window, such images would appear truncated, negatively impacting the
With the on-sky data taken with the CHEC-M camera we proved
both the technical functionality of the camera and the existence of
a data calibration and analysis chain, both aspects being crucial for
the camera to be used as an IACT camera. Both campaigns helped to
J. Zorn et al.
Nuclear Inst. and Methods in Physics Research, A 904 (2018) 44–63
– a mean baseline drift with temperature of 0.8 ADC/◦ C
is observed and is mitigated by maintaining a constant
camera temperature within 1 ◦ C or by taking regular 30 s
pedestal calibration runs.
verify interfaces and to improve operation procedures. Furthermore, the
regular operation was used for understanding the system stability and
reliability. In total, a few hundred meaningful Cherenkov and direct EAS
particle events were recorded and analysed.
The on-telescope campaigns have not only provided a useful test-bed
to assess operational and maintenance procedures, but have also – for
the first time – demonstrated the use of Schwarzschild–Couder optics
to collect atmospheric Cherenkov light. Data taken during the two
campaigns have proven useful in the development of the data analysis
chain and in understanding the levels of calibration that will be required
for CTA.
Despite the level of success achieved with the prototype, CHECM does not meet all CTA performance requirements, with the nonuniformity in gain and the trigger noise incurred with sampling enabled
being of greatest concern. CHEC-S, a second full-camera prototype based
on SiPMs, is currently being commissioned and tested, and will address
the limiting factors in the CHEC-M performance by design.
The use of SiPMs allows gain measurements to be made easily for
a range of bias voltages and input illumination levels (and even in the
absence of light from the dark counts intrinsic to the SiPM) thereby
improving calibration and charge reconstruction accuracy. The SiPM
gain spread between pixels in a camera module is intrinsically smaller
than with MAPMs, and the bias voltage is adjustable per superpixel,
allowing gain matching to much higher precision than in CHEC-M. The
gain of SiPMs is temperature sensitive, and for the devices used in
CHEC-S will drop by approximately 10%–20% over a 10 ◦ C increase. A
liquid-cooled focal plane plate will stabilise the temperature to within
±1 ◦ C over time scales for which the gain may easily be re-measured
in-situ. The average detection efficiency in the focal plane will also
increase, due to improved photo-detection efficiency, better angular
response,18 and a reduced level of dead space between photodetectors.
NSB rates are expected to be higher than in the MAPM case, due to the
different wavelength dependence of the SiPM response, however, this
background increase should be compensated for by the improvement
in efficiency for signal photons. Dark count rates from the SiPMs at
the nominal operating gain and temperature have been measured to be
less than 20% of the expected dark sky NSB rate, ensuring a negligible
impact on performance.
The FEE of CHEC-S will also see a substantial upgrade from CHEC-M.
Due to the undesirable coupling between sampling and triggering in the
TARGET 5 ASICs, these functionalities have been split into two separate
ASICs. The first ASIC, T5TEA, provides triggering based on the same
concept as TARGET 5, with a sensitivity reaching the single p.e. level
and a trigger noise of 0.25 p.e. for the CHEC-S gain. The second ASIC,
TARGET C, performs sampling and digitisation, with a ∼70% larger
dynamic range and with an improvement in charge resolution by a factor
> 2 with respect to TARGET 5 [33]. The operational requirement for an
80 μs hold-off time has therefore been removed, allowing operation dead
time free at the required event trigger rate of 600 Hz.
In the final CHEC design, the absolute timing will be provided by
a unified clock and trigger timestamping board. The board acts as an
interface between the camera and the CTA timing system and is based
on White Rabbit technology [25]. It provides clock signals to the camera
with the required precision that are phase-locked to the central master
clock. In return, it adds absolute timestamps to the camera events.
Furthermore, this board will have the capability to trigger the backplane
and LED calibration flashers (synchronised with the internal clock), thus
replacing the need of an external trigger device. A prototype of this
board will be integrated into CHEC-S.
Once prototyping is complete, we plan to construct and deploy
three CHEC cameras on the southern-hemisphere CTA site during a
pre-production phase. During the production phase of CTA we aim to
7. Summary and outlook
CHEC-M is an invaluable step towards a reliable and highperformance product for CTA. Regular operation of CHEC-M has shown
that the camera control and data acquisition using the software CHECInterface and the calibration and waveform processing chain are robust
and reliable. Intensive lab tests led to a detailed characterisation of the
camera performance as well as a detailed understanding of the factors
limiting the performance, thus being a critical input to the design of the
next camera prototype CHEC-S, see below). Main results are:
∙ CHEC-M can be read out at an efficiency of 95% at a mean
random rate of 600 Hz. The efficiency is expected to be 100%
at this rate with the next TARGET-5 ASIC generation used in
∙ While commissioning and testing of CHEC-M, 128 TARGET ASICs
have been tested and tuned simultaneously confirming the results
obtained in single TARGET ASIC tests (cf. [24]).
∙ Even after gain matching, the spread in gain between pixels is
∼30% — the limiting factor in trigger threshold uniformity. This
is due to the fact that the HV can only be set individually for
each MAPM but not for each pixel — a fundamental feature of
the MAPM design. The gain spread is reduced significantly when
using SiPMs as photosensors (which is the case in CHEC-S).
∙ Camera trigger thresholds are characterised by 512 pairs of two
ASIC parameters for each threshold which can be determined in
lab measurements with a laser. In this way, five different threshold sets were defined for on-site tests between ∼2 and ∼170 p.e.,
intermediate thresholds can be identified by interpolation. A
trigger rate scan over these settings can be used to identify the
camera operating point on site.
∙ Reading out TARGET-5 modules leads to additional, false, triggers resulting in increased dead-time and the need of an increased trigger threshold, both issues being addressed by decoupling sampling and triggering into two separate ASICs in the next
TARGET module generation.
∙ The pulse shape characteristics (FWHM and 10%–90% risetime)
fulfil the requirements for trigger performance optimisation (5–
10 ns and 2–6 ns, respectively).
∙ The time resolution between different pixels hit simultaneously
by the same laser flash is better than 1 ns for illumination levels
>6 p.e. and thus also fulfils the requirements.
∙ The crosstalk reaches a maximum of 6% between neighbouring
pixels, affecting charge resolution, trigger efficiency, and camera
image reconstruction.
∙ The dynamic range of the signal recording chain (MAPM, preamplifier, and TARGET module) covers a range from the subp.e. level to ∼1000 p.e. at the highest possible HV of 1100 V
and can be shifted towards higher signal amplitudes (factor ∼6)
by reducing the gain.
∙ The LED calibration flashers were shown to be appropriate
devices for regular camera calibration and monitoring in terms of
absolute gain and dynamic range determination. For the future,
a slight change in their design is considered to avoid different
temperature dependencies and to improve the predictability of
absolute and relative brightnesses of different LED patterns.
∙ The camera and temperature stability were assessed showing that
– the camera temperature can be controlled and (if required)
kept at a constant level by adapting the chiller temperature,
– the camera warm-up time is of the order of 1 h, and
MAPMs need a protective glass worsening their angular response compared
to SiPMs.
J. Zorn et al.
Nuclear Inst. and Methods in Physics Research, A 904 (2018) 44–63
provide cameras for a significant fraction of the 70 baseline SSTs. It is
expected that the majority of components used in CHEC-S will also be
used in the final production design of CHEC with the exception of the
photosensors. SiPM technology is rapidly evolving and the latest devices
offer significant performance improvements compared to the SiPMs used
in CHEC-S, including increased photo-detection efficiency, lower optical
crosstalk and a reduced dependency of the gain on temperature [34–36].
Additionally, there may be performance advantages associated with an
enlarged FoV that may be obtained by using 7 mm rather than 6 mm
pixels. Laboratory tests of the latest SiPMs and simulations with different
pixel sizes are ongoing.
[14] C. Bigongiari, The MAGIC Telescope, in: Proceedings, 2005 Europhysics Conference
on High Energy Physics (EPS-HEP 2005): Lisbon, Portugal, July 21–27, 2005, PoS
HEP2005 (2006) 020.
[15] K. Schwarzschild, Untersuchungen zur geometrischen Optik I, Astron. Mitt. Univ.Sternwarte Goett. 9 (1905) 1.
[16] A. Couder, 183 (II), 1276, Tech. Rep., C.R. Acad. Sci. Paris, 1926.
[17] V. Vassiliev, S. Fegan, P. Brousseau, Wide field aplanatic two-mirror telescopes for
ground-based -ray astronomy, Astropart. Phys. 28 (2007) 10–27.
[18] P. Laporte, et al., SST-GATE: an innovative telescope for very high energy astronomy,
in: Ground-Based and Airborne Telescopes IV, in: Proceedings of the SPIE, vol. 8444,
2012, p. 84443A.
[19] J.L. Dournaux, et al., Operating performance of the gamma-ray Cherenkov telescope:
An end-to-end Schwarzschild-Couder telescope prototype for the Cherenkov Telescope Array, in: 14th Vienna Conference on Instrumentation (VCI 2016) Vienna,
Austria, February 15–19, 2016, Nucl. Instrum. Meth. A845 (2017) 355–358. http:
[20] S. Vercellone, et al., The ASTRI Project: An innovative prototype for a Cherenkov
dual-mirror small-telescope, in: AAS/High Energy Astrophysics Division, in: AAS/
High Energy Astrophysics Division, vol. 13, 2013, p. 123.31.
[21] K. Bechtol, et al., TARGET: A multi-channel digitizer chip for very-high-energy
gamma-ray telescopes, Astropart. Phys. 36 (2012) 156–165.
[22] K. Bernlöhr, et al., Monte Carlo design studies for the Cherenkov Telescope Array,
Astropart. Phys. 43 (2013) 171–188.
[23] A. De Franco, et al., The first GCT camera for the Cherenkov Telescope Array,
in: Proceedings, 34th International Cosmic Ray Conference (ICRC 2015): The Hague,
the Netherlands, July 30–August 6, 2015, PoS ICRC2015 (2016) 1015.
[24] A. Albert, et al., TARGET 5: A new multi-channel digitizer with triggering capabilities
for gamma-ray atmospheric Cherenkov telescopes, Astropart. Phys. 92 (2017) 49–61.
[25] J. Serrano, et al., The white rabbit project, in: Proceedings of ICALEPCS TUC004,
Kobe, Japan, 2009.
[26] A. Brown, et al., Flasher and muon-based calibration of the GCT telescopes proposed
for the Cherenkov Telescope Array, in: Proceedings, 34th International Cosmic Ray
Conference (ICRC 2015): The Hague, the Netherlands, July 30-August 6, 2015, PoS
ICRC2015 (2016) 934.
[27] D. Hanna, Calibration Techniques for VERITAS, Int. Cosmic Ray Conf. 3 (2008)
[28] W.D. Pence, et al., Definition of the flexible image transport system (fits), version
3.0, Astron. Astrophys. 524 (2010) A42.
[29] J.D. Case, et al., Simple network management protocol (SNMP), RFC 1157 (1990).
[30] CTA Experimental Pipeline Framework (ctapipe), CTA, 2018, https://cta-observato
[31] F. Werner, et al., Performance verification of the FlashCam prototype camera for the
Cherenkov Telescope Array, in: Proceedings, 9th International Workshop on Ring
Imaging Cherenkov Detectors (RICH 2016): Bled, Slovenia, September 5–9, 2016,
Nucl. Instrum. Meth. A876 (2017) 31–34.
[32] J.J. Watson, et al., Inauguration and First Light of the GCT-M Prototype for the
Cherenkov Telescope Array, in: Proceedings, 6th International Symposium on HighEnergy Gamma-Ray Astronomy (Gamma 2016): Heidelberg, Germany, July 11–
15, 2016, AIP Conf. Proc. 1792 (1) (2017) 080006.
[33] S. Funk, et al., TARGET: A digitizing and trigger ASIC for the Cherenkov telescope
array, in: 6th International Symposium on High Energy Gamma-Ray Astronomy,
in: American Institute of Physics Conference Series, vol. 1792, 2017, 080012. http:
[34] H. Tajima, et al., Evaluation of silicon photomultipliers for the GCT and ASTRI
cameras of CTA, in: 8th International Conference on New Developments in Photodetection, 2017.
[35] A.N. Otte, et al., Characterization of three high efficiency and blue sensitive silicon
photomultipliers, Nucl. Instrum. Meth. A846 (2017) 106–125.
[36] A. Asano, et al., Evaluation of silicon photomultipliers for dual-mirror small-sized
telescopes of Cherenkov Telescope Array, Nucl. Instrum. Methods Phys. Res. A
This work was conducted in the context of the CTA GCT project.
We gratefully acknowledge financial support from the agencies and
organisations listed here:
Furthermore, we thank the Paris Observatory as well as the DT-INSU
for their support during the on-sky campaigns in Meudon. This study
was also supported by JSPS KAKENHI Grant Numbers JP17H04838,
JP25610040, JP15H02086, and JP23244051. A. Okumura was supported by a Grant-in-Aid for JSPS Fellows.
This paper has gone through internal review by the CTA Consortium.
[1] A.U. Abeysekara, et al., Observation of the Crab Nebula with the HAWC gammaray observatory, Astrophys. J. 843 (1) (2017) 39.
[2] G. Di Sciascio, The LHAASO experiment: from Gamma-Ray Astronomy to Cosmic
Rays, in: Proceedings, 9th Cosmic Ray International Seminar: The Status and the
Future of the UHE Cosmic Ray Physics in the Post LHC Era (CRIS 2015): Gallipoli,
Italy, September 14–16, 2015, Nucl. Part. Phys. Proc. 279–281 (2016) 166–173.
[3] M. Actis, et al., Design concepts for the Cherenkov Telescope Array CTA: an advanced
facility for ground-based high-energy gamma-ray astronomy, Exp. Astron. 32 (2011)
[4] G.P. Rowell, et al., TenTen: A new IACT array for multi-TeV -ray astronomy,
Nucl. Instrum. Methods Phys. Res. A 588 (2008) 48–51.
[5] J.A. Hinton, et al., A new era in gamma-ray astronomy with the Cherenkov telescope
array, Astropart. Phys. 43 (2013) 1–2.
[6] R. Cornils, et al., The optical system of the H.E.S.S. imaging atmospheric Cherenkov
telescopes. Part II: mirror alignment and point spread function, Astropart. Phys. 20
(2003) 129–143.
[7] J.A. Hinton, The HESS Collaboration, The status of the HESS project, New Astron.
Rev. 48 (2004) 331–337.
[8] I. de la Calle Pérez, S.D. Biller, Extending the sensitivity of air Čerenkov telescopes,
Astropart. Phys. 26 (2006) 69–90.
[9] A. Daum, et al., First results on the performance of the HEGRA IACT array, Astropart.
Phys. 8 (1997) 1–11.
[10] A. Akhperjanian, The optical layout of the HEGRA Cherenkov telescopes, Exp.
Astron. 8 (2) (1997) 135–152.
[11] J.M. Davies, E.S. Cotton, Design of the quartermaster solar furnace, Sol. Energy 1
(1957) 16–22.
[12] T.C. Weekes, et al., VERITAS: the very energetic radiation imaging telescope array
system, Astropart. Phys. 17 (2002) 221–243.
[13] J. Holder, VERITAS: Status and highlights, Int. Cosmic Ray Conf. 12 (2011) 137.
Без категории
Размер файла
4 549 Кб
078, niman, 2018
Пожаловаться на содержимое документа